Eli's AI Daily 30 April 2026

Five stories. Twenty headlines. Everything that matters in AI today.

Good morning. Last night four of the world's biggest companies reported earnings within 80 seconds of each other. Alphabet's AI bet paid off spectacularly. Meta's did not, yet. Elon Musk called himself a fool in a federal courtroom. The White House quietly moved to bring Anthropic back in from the cold. And OpenAI landed on Amazon Bedrock exactly 48 hours after its exclusivity deal with Microsoft was torn up. Welcome to Thursday.

Story 01

Big Tech earnings: Alphabet's AI bet paid off. Meta's has not. Yet.

Four companies reported within 80 seconds of each other last night. The verdict on AI spending is in, and it is not uniform.

What happened

Alphabet, Meta, Microsoft, and Amazon all reported Q1 2026 earnings last night within a span of two minutes after the bell. Alphabet was the standout. Revenue hit $109.9 billion, up 22%, with net profit up 81% to $62.6 billion. Google Cloud grew 63% to $20 billion, its fastest growth in years. Sundar Pichai called it the company's "strongest quarter ever" for consumer AI plans, with 350 million paid subscriptions. Meta beat on revenue at $56.3 billion but missed on forward guidance, flagging flat Q2 growth and raising its full-year AI capex outlook to $125-145 billion. Its shares fell more than 5% after hours. Microsoft beat estimates with $82.9 billion in revenue. Amazon reported $181.5 billion in revenue, with AWS at $37.6 billion, both above consensus. Microsoft and Amazon shares fell around 3% despite the beats, as investors found the guidance insufficiently bullish.

Why it matters

The divergence between Alphabet and Meta is the most important signal from last night. Both companies are spending colossal sums on AI. Alphabet is seeing the return already, primarily through Google Cloud's enterprise AI growth and Search AI features. Meta is not seeing it yet, and its own guidance suggests it does not expect to in Q2. The market is now in a position where it can compare AI strategies across comparable scales of investment and draw real conclusions. Alphabet's 63% Cloud growth suggests that enterprises are actually buying AI infrastructure. Meta's flat Q2 guidance suggests that AI-driven advertising uplift is not yet materialising at the pace the market priced in. These are two different AI business models being tested at the same time. One is passing. One is on notice.

The question nobody's asking

Microsoft and Amazon both beat estimates and both fell 3% after hours. At what point does beating Wall Street's expectations simply mean the expectations were set too low?

Story 02

Musk calls himself a fool in court. His cross-examination continues today.

Day three of the Musk v. OpenAI trial produced the most quotable testimony of the week, and it is not over yet.

What happened

Elon Musk returned to the stand yesterday for his second day of testimony in his lawsuit against OpenAI and Sam Altman in Oakland, California. Under questioning from his own lawyer, Musk said he was "a fool" for putting money into OpenAI, and that OpenAI's leaders were "looting the nonprofit." He described his faith in the company in three phases: initial enthusiasm, growing uncertainty, and a final period beginning in late 2022 when he felt "they were betraying" everything OpenAI was built on. Under cross-examination from OpenAI's attorney William Savitt, a heated exchange broke out. Savitt pressed Musk on whether he had read a term sheet from several years ago. Musk said Savitt's questions were "definitionally complex, not simple. It is a lie to say they are simple." The judge intervened multiple times. Musk's testimony will continue today. Greg Brockman has been given 48-hour notice to testify, meaning he could take the stand as early as tomorrow.

Why it matters

OpenAI's cross-examination strategy is becoming clear. Rather than defending the for-profit conversion on its merits, Savitt is systematically attacking Musk's motives. The argument is not "we did nothing wrong" but "Musk wanted control, didn't get it, and launched xAI instead." That reframe is effective because it sidesteps the diary, the founding documents, and the nonprofit promises. It asks the jury to decide what kind of person Musk is rather than what kind of company OpenAI became. The cross-examination has also surfaced the fact that Musk pledged $1 billion to OpenAI but contributed only $38 million, and that he left after losing a leadership dispute. These are not flattering details for a plaintiff who says he was betrayed by others.

The question nobody's asking

Musk said his own reputation was worth more than the $38 million he contributed. Is that a legal argument, a negotiating position, or just something that sounded better out loud than on paper?

Story 03

The White House is quietly trying to bring Anthropic back in from the cold

Anthropic refused the Pentagon's terms. Got blacklisted. Now the White House wants to reverse the standoff. Someone blinked.

What happened

The White House is drafting guidance that would allow federal agencies to work with Anthropic again, including access to the company's unreleased model Mythos, according to reporting from Axios published today. The move comes after the Pentagon designated Anthropic a "supply chain risk" following Anthropic's refusal to accept broad AI access terms for classified government work. Anthropic is currently fighting that designation in federal court. The White House guidance would create a pathway for agencies to onboard Anthropic's models while the legal dispute is resolved, effectively working around the Pentagon's blacklisting without formally overturning it. The administration separately confirmed it views Mythos as a strategically significant model for national security applications.

Why it matters

Read yesterday's story about Google signing the Pentagon deal alongside today's story about the White House trying to restore Anthropic, and a clear picture emerges. The government needs Anthropic's models. Anthropic's refusal to accept unrestricted classified use was not a deal-breaker for the Pentagon politically, it was a negotiating position that the administration is now working to resolve. The distinction matters because it changes how you read Google's decision to sign. Google did not sign because it had no principles. It signed because the alternative was exclusion from a market the government will eventually bring everyone back into anyway. Anthropic played a longer game and may be winning it. Whether the White House pathway preserves Anthropic's original red lines or quietly abandons them will determine whether holding the line was worth it.

The question nobody's asking

If the White House creates a pathway that bypasses the Pentagon's blacklist without changing the underlying terms, has Anthropic won anything, or just been given a more polite version of the same deal?

Story 04

OpenAI lands on Amazon Bedrock. 48 hours after ending exclusivity with Microsoft.

The speed of this move tells you everything about how long it had been planned.

What happened

OpenAI models including GPT-5.5 are now available on Amazon Bedrock, announced today as part of the AWS "What's Next" event. Alongside the model availability, AWS launched Amazon Bedrock Managed Agents powered by OpenAI in limited preview, and OpenAI's Codex is now available on Bedrock for enterprise software development workflows. The partnership gives enterprises access to OpenAI's frontier models within AWS's existing security, compliance, and identity infrastructure, removing the need to build separate pipelines. More than 4 million people now use Codex weekly. The Bedrock Managed Agents service handles orchestration, tool use, and governance, offering a faster path from prototype to production for agentic deployments. AWS CEO Matt Garman described it as "frontier intelligence on the infrastructure enterprises already trust."

Why it matters

The 48-hour gap between the exclusivity announcement and the Bedrock launch is not a coincidence. This was clearly negotiated in parallel with the Microsoft restructuring, which means OpenAI has been executing a multi-cloud distribution strategy for months while publicly maintaining an exclusive Azure relationship. That is a significant shift in how OpenAI goes to market. Azure built its enterprise AI business substantially on the back of OpenAI exclusivity. That moat is now gone. Google Cloud has Gemini. AWS now has GPT-5.5, Codex, and Managed Agents. OpenAI has effectively become a model provider to all three major clouds simultaneously, which dramatically increases its distribution but also its exposure to commoditisation. When your product is available everywhere, differentiation becomes the only thing that matters.

The question nobody's asking

OpenAI is now available on AWS, Azure, and Google Cloud. At what point does being everywhere become the same thing as being a commodity?

Story 05

The EU AI Act August deadline is still in force. Brussels walked out without a deal.

A 12-hour trilogue in Strasbourg ended without agreement on the Omnibus. If you were waiting for a delay, stop waiting.

What happened

The European Commission, Council and Parliament walked out of a 12-hour trilogue session in Strasbourg on April 28 without reaching agreement on the AI Act Omnibus package. The sticking point was a single file: the conformity-assessment relationship between the AI Act and existing sectoral rules for medical devices and machinery. No deal means the original AI Act timeline remains legally in force. The August 2, 2026 deadline for high-risk AI systems now applies in full. Any AI used in recruitment, screening, performance assessment, or termination is classified as high-risk under Annex III. The next trilogue is expected around May 13 under the Cypriot Presidency. Penalties for prohibited practices reach 35 million euros or 7% of worldwide annual turnover, whichever is higher.

Why it matters

Hundreds of companies building or deploying AI in Europe had been quietly banking on the Omnibus softening the August deadline. That expectation is now gone. The collapse of the trilogue is not a minor procedural setback. It means that any AI system used in hiring, firing, or performance management that touches EU employees or candidates must meet the full high-risk compliance requirements in 14 weeks. That includes audit trails, human oversight mechanisms, data governance documentation, and conformity assessments. Many companies are not ready. The next trilogue on May 13 is unlikely to produce a deal in time to meaningfully change the August 2 position. If you are building AI hiring tools and you were waiting for the Omnibus, you have run out of time to wait.

The question nobody's asking

The companies that lobbied hardest for the Omnibus softening are the same companies deploying AI hiring tools at scale. If the deadline holds and enforcement begins in August, which regulator moves first, and which company becomes the example?

Two tools worth your time

Amazon Bedrock Managed Agents (OpenAI) — Live today in limited preview. If you are building enterprise agentic workflows and already run on AWS infrastructure, this is the fastest path from prototype to production available right now. Sign up for preview access via the AWS console.

EU AI Act Annex III checker — Before August 2, run your AI products through the European Commission's online self-assessment tool to confirm whether they fall under high-risk classification. It takes 20 minutes and could save considerably more than that in legal fees.

What else happened in AI today

Twenty things worth knowing. One line each.

01 Alphabet raised its 2026 AI capex guidance and said 2027 spending will "significantly increase" — its CFO's words, not an analyst's projection.
02 Google Cloud's backlog nearly doubled quarter on quarter to over $460 billion, the largest enterprise AI pipeline ever reported by any cloud provider.
03 Meta raised its full-year AI capex guidance to $125-145 billion, up from its prior range, on the same call where it guided flat Q2 revenue growth.
04 Amazon Web Services revenue came in at $37.6 billion for Q1, beating estimates, with AI workloads cited as the primary growth driver for the third consecutive quarter.
05 Rogo, the agentic AI platform for investment banking, closed a $160 million Series D led by Kleiner Perkins, bringing total funding north of $300 million, with over 35,000 finance professionals now using its agent Felix.
06 Greg Brockman has been given 48-hour notice to testify in the Musk v. OpenAI trial, meaning he could take the stand as early as Friday.
07 The US Senate is expanding its AI safety probe into major labs including OpenAI, adding a new regulatory pressure track running parallel to the Oakland trial.
08 Amazon launched Amazon Quick, a new AI work assistant with a desktop app and free tier, competing directly with Microsoft Copilot and Google Workspace AI.
09 Google's Gemini Enterprise saw 40% quarter-on-quarter growth in paid monthly active users, its fastest adoption rate since launch.
10 OpenAI released Privacy Filter as an open-source tool, achieving state-of-the-art performance on the PII-Masking-300k benchmark for protecting personally identifiable information in AI pipelines.
11 72% of US physicians now use AI in clinical practice, up from 48% last year, according to a 2026 American Medical Association survey published today.
12 Apple reports Q2 2026 earnings today at 5pm ET, with investors watching the Gemini-powered Siri roadmap and guidance from incoming CEO John Ternus more closely than the revenue number.
13 Amazon Connect is expanding into four agentic AI solutions covering supply chains, hiring, customer experience, and healthcare, all announced at the AWS What's Next event today.
14 Google's Gemini app now has 350 million paid subscribers globally, making it the fastest-growing AI consumer subscription product in history.
15 Microsoft spent nearly $30 billion on AI infrastructure in a single quarter, up 89% year on year, the largest single-quarter AI capex number ever reported by any company.
16 OpenAI's Codex now has over 4 million weekly active users, more than doubling since the start of the year, according to figures shared at the AWS Bedrock launch today.
17 The EU AI Act next trilogue session is expected around May 13 under the Cypriot Presidency, though a deal in time to change the August 2 high-risk deadline is considered unlikely by Brussels insiders.
18 Anthropic's unreleased model Mythos is now confirmed as a national security priority by the White House, the first time any government has publicly named an unreleased AI model as strategically significant.
19 Google Search hit an all-time high in queries this quarter, with Sundar Pichai attributing the growth directly to AI-powered search features driving more usage, not less.
20 Judge Yvonne Gonzalez Rogers intervened multiple times during Musk's cross-examination yesterday, a detail that trial watchers say signals she is keeping a close eye on the credibility of testimony from both sides.
That is your Thursday. Alphabet proved that AI infrastructure spending can translate into real revenue. Meta proved it does not happen automatically. Musk called himself a fool on the stand and kept fighting anyway. The White House is cleaning up a mess of its own making with Anthropic. And OpenAI is now available on every major cloud platform simultaneously, 48 hours after ending exclusivity. The AI industry is moving so fast that yesterday's competitive moat is today's press release. Have a good weekend.

— Eli

Reply

Avatar

or to participate

Keep Reading

background